Fires in buildings can escalate rapidly, posing significant risks to both life and property. Consequently, early fire detection (FD) and timely warnings are crucial. Smoke and FDs are the current technology used for indoor detection. These detectors do have a few drawbacks, though, both when a fire is starting and when it is spreading. To overcome these issues, a novel FIRE pixel detection Building Environment via Internet of Things (IOT) based vision sensor (FIRE-BET) has been proposed for FD in smart buildings based on computer vision (CV). The approach that is being suggested breaks the procedure down into three steps: the first is based on a Gaussian probability distribution (GPD), the second is a spatio-temporal graph-based approach, and the third is based on temporal variation. These three stages are typically utilized to address a variety of problems, including Gaussian noise in frames and various video resolutions. The experiments' findings show outstanding detection accuracy for video frames in a variety of lighting situations and are resistant to noise like smoke. The proposed work achieves the high accuracy range of 91.83%, TPR of 92.77%, and less FPR of 16.66%.